IncudoCAD
CTO
Jon
The builder.
Owns the CAD engine: Rust + C++ core, geometry math, and the ML models that handle what pure math can't. Previously ML Engineer @ J6 & Cyber Defence Directorate, Senior SWE @ CloudQuery.
Engineer in MeshLab again. We can tell.Scan to CAD. Automated.
Scan to CAD. Automated.
One API call.
We take the manual, repetitive scan-to-CAD labor off your engineers: the 20–40 minutes per file in MeshLab, Geomagic, or Polyworks. They stay on design, custom problems, and customer-facing work.
Reconstruction algorithms + AI, dynamic to your use case.
INPUT · MESH (.stl)HEALING · POST /v1/healOUTPUT · CAD-NATIVE (.step)flanged hub
§02 Problem
One intake. Three processes. One broken file stalls the queue.
Mixed-process shops handle STL, STEP, and scans through the same intake desk. A non-manifold mesh stalls 3DP scheduling. A faceted STEP stalls CAM. A raw scan stalls the next handoff. Different processes, same blocker.
Your ops engineer ends up in three different tools (MeshLab for the STL, Geomagic for the scan, SOLIDWORKS for the STEP), and the shop's throughput is rate-limited by whichever file landed worst that morning.
01IntakeSTL, STEP, and scans all hit the same desk
02Broken fileNon-manifold edges, holes, faceted STEP, raw scan noise
03Ops engineer juggles three repair toolsMeshLab / Geomagic / SOLIDWORKS, per file type
04Downstream blockedWhatever process the file was destined for waits its turn
Scan-to-CAD dies in cleanup.
At a scan-service or RE bureau, every job ends with the same cleanup loop: close holes, fix non-manifold edges, force the geometry into something a CAD kernel will accept: Geomagic, MeshLab, manual stitching. That loop is where scan-to-CAD-Native-STEP work goes to die.
The cost is silent and persistent: senior engineers babysitting cleanup queues instead of designing, scans rejected downstream because the file isn't watertight, jobs delayed because the mesh→STEP conversion failed in three places no one wants to find.
01Scan arrives rawSTL, OBJ, PLY, straight off the scanner, unprocessed.
02Mesh is unusableHoles, non-manifold edges, no watertight topology.
03Engineer babysits cleanupHours in Geomagic or MeshLab, per scan.
04STEP export failsmesh→CAD conversion breaks in ways no one wants to debug.
Files arrive broken. CAM can't tool-path mesh.
Customer files come in every format (STL, OBJ, 3MF, STEP, IGES), and most are broken in ways that block CAM and quoting. Customer scans now arrive routinely alongside the STEP/IGES inbox, and CAM needs tool-pathable geometry, not faceted mesh. Faceted STEP that won't tool-path. IGES that's lost its topology. Meshes with holes you don't see until the toolpath fails.
That's setup time you can't bill, quote turnaround you can't compress, and rework you can't predict. A tax paid on every job before the spindle moves.
01Customer uploads a fileSTL, OBJ, 3MF, STEP, or IGES, from their CAD tool.
02File is broken on arrivalFaceted STEP, corrupt IGES, meshes with holes you won't see until toolpath runs.
03Engineer repairs manuallyHours of cleanup before quoting can even start.
04Quote stalls or ships lateSetup time you can't bill. Rework you can't predict.
§03 How it works
Three steps. One endpoint.
POST your file, get geometry back in the format you asked for. No license negotiation.
01 / POST your file
STL, OBJ, 3MF in.
Upload directly to our edge. No CLI, no SDK install. One HTTP call.
02 / We process
Three modes, same endpoint.
Smart single-pass by default handles the routine 90%. Iterative or Neural (AI) when the geometry needs more. Caller picks, per call.
03 / Get CAD-Native STEP
Or any format you ask for.
STEP, IGES, STL, OBJ, or 3MF out. Caller picks. Confidence score on every response. Below threshold → flagged, not shipped.
// Requestoctet-stream
POST /v1/heal
Content-Type: application/octet-stream
X-IncudoCAD-Key: ic_live_•••
Accept-Output: step
// {file bytes: STL, OBJ, or 3MF}
// Response 200 OKapplication/json
{
"status": "healed",
"confidence_score": 0.94,
"output_format": "step",
"download_url": "https://..."
}
§04 Proof
Measured, not marketed.
We publish benchmarks against public CAD corpora when they're real, not before.
Benchmarks coming very soon, published when they're real.
§06 How we're different
One platform. Three layers.
Healing is the entry pain. Format interop is the breadth. The mesh → CAD-Native STEP step is the one nobody else automates at this price.
01 / The hard one
Mesh → CAD-Native STEP, automated.
The hard one nobody else automates at this price. SOLIDWORKS 2025 has manual mesh-to-CAD conversion in the desktop app. You click through segmentation, one file at a time. Spatial sells it as an SDK add-on. IncudoCAD does it headless, in one API call, at pipeline speed. Built for shops where this conversion is a daily problem, not an occasional one.
02 / Breadth
One API. Five formats. Both directions.
STL, OBJ, 3MF, STEP, and IGES: read and write, healed in the same call. Spatial and Tech Soft 3D license C++ SDKs to software vendors for five- and six-figure annual deals; you don't need to build an app around a library to bridge formats. One endpoint. Pay per call. No license negotiation.
03 / Trust
Trust is the product.
Every response includes a confidence score. If we can't heal a file above threshold, the API flags it for manual review instead of silently shipping broken geometry. Whatever's downstream (CAM, quoting, inspection, redesign) never gets a bad file.
§07 Pricing
Start free. Pay for what you use.
Three pipeline modes through the same endpoint. Default is an algorithmic single pass. Iterative and Neural are opt-in.
Tier 01
Smart
- Trigger
- Default · CPU
- Use case
- Standard meshes: topological errors, non-manifold edges, hole filling. Simple prismatic parts.
- Engine
- Algorithmic single-pass: Rust + OpenCascade core.
Tier 02
Iterative
- Trigger
- Opt-in · CPU
- Use case
- Complex parts that resist a single pass: multiple passes refine the reconstruction with parameter tuning until output hits tolerance.
- Engine
- Algorithmic multi-pass: same core, iterative refinement loop.
Tier 03
Neural
- Trigger
- Opt-in · GPU
- Use case
- Scan-captured meshes, disastrous input, geometry that the algorithmic tiers can't recover.
- Engine
- Serverless GPU + AI-driven neural reconstruction: dedicated ML models for segmentation and surface fitting.
Start free while we publish benchmarks. Early-access pricing for shops we onboard now. Talk to us and we'll size it to your volume.
Start a free trial, talk to us §09 Team
Who's building this.
Two engineers. Alumni of Israel's top tech units. The ones writing the code answer your support emails.
CEO
Yotam
The driver.
Leads business, product & infrastructure. Former Head of DevOps @ SignalPET ($15M+ ARR) & Technical Product Manager @ 8200.
§10 FAQ
Questions you'd ask in a sales call.
Same answers, no calendar invite. If yours isn't here, mail team@incudocad.com.
Every call returns a confidence score in [0, 1]. Files below threshold are flagged for manual review rather than silently returned, so whatever's downstream (CAM, quoting, inspection, redesign) never gets bad geometry. [benchmark: accuracy across Thingi10K/ABC, pending].
Start free. Pay for what you use when you're ready. No seats, no SDK license, no minimums. Three tiers: Smart (default deterministic), Iterative (opt-in deterministic multi-pass), and Neural (opt-in GPU ML). Early-access pricing for shops we onboard now. Talk to us.
SOLIDWORKS 2025 ships manual mesh-to-CAD conversion in its desktop GUI. You segment facets by hand, file by file. IncudoCAD is the API version: one endpoint, headless, wires into your existing pipeline.
Geomagic Design X is desktop reverse-engineering software: an engineer segments the scan and rebuilds each part by hand. IncudoCAD automates that. Our AI handles the messy conversions that otherwise need a human in the loop, and returns CAD-Native STEP from a single call. They need an engineer per part; we need one click.
Spatial and Tech Soft 3D sell C++ SDKs to software vendors who embed them in an app. IncudoCAD is a hosted REST API you call directly from your shop's scripts, intake tool, or pipeline. You don't need to build an app around us.
Files run on dedicated infrastructure and are encrypted in transit and at rest. We don't sell your data and we don't share it for advertising. Full retention, encryption, and access controls are in our data handling policy. Happy to sign an NDA for specific workflows.
Yes. Our Web UI is live: same engine as the API, same Smart / Iterative / Neural tiers. Access is currently gated; book a call and we'll set you up with a key. When you're ready to automate intake, switch to the API.
§11 Free demo, on a call
Get your engineer's time back.
Drop your contact + a note on your scan workflow. We'll book a call to show how much engineer time the API frees up. Per week, per month.